The Outlook: Today's Disinformation Machines Make Digital Noise Like Humans, Too

The Lookout tells us what happened. The Outlook asks what it means.

The data comes from FC26, Michigan United's election early-warning program, which archives public election posts through Open Measures. Counts only. No accounts are named.

Today’s Lookout shows several recurring families of rumors and questionable claims surrounding Michigan’s 2026 election.

The largest is “Doubt the Rolls.”

During the past 30 days, we identified:

• 11 separate waves
• A typical gap of about one day between waves
• 18 accounts appearing in more than one wave

Another recurring family, “Doubt the Referee,” produced:

• Five waves
• A typical gap of about four days
• 14 accounts appearing in more than one wave

“Doubt the Count” produced six waves. “Scare Them Off” has appeared three times recently as Michigan enters early voting.

Those are the observations.

The harder question is what they mean.

Because election misinformation is no longer only a content problem.

It is increasingly a problem of behavior, networks and amplification.

And one of the assumptions we have relied on for years is becoming much less useful:

Humans create noise. Machines create regularity.

That distinction is breaking down.

Machines can create regularity.

Machines can also create noise.

And that means the noise itself can become camouflage.

The old bot model was easier

The easiest automated network to identify looks something like this:

One account posts a claim.

Ten accounts repeat it immediately.

The language is identical.

The timing is regular.

The same accounts keep interacting.

That type of activity produces an obvious machine-like signature.

But sophisticated automation does not have to behave that way.

It can vary the timing.

It can rewrite the message.

It can change the tone.

It can mix political content with sports, weather, entertainment or everyday conversation.

One account can make an accusation.

Another can ask a question.

Another can sound skeptical.

Another can post a meme.

Another can appear to disagree.

The surface becomes messy.

The underlying purpose may remain the same.

What if the messiness is intentional?

Consider five hypothetical posts about voter rolls:

“The voter rolls are corrupt.”

“I’m not saying there is fraud, but these numbers make no sense.”

“Has anyone actually explained what is happening with these registrations?”

“People are exaggerating, but election officials really should answer some questions.”

“I just want transparency.”

They do not look identical.

They do not even take exactly the same position.

But they can still produce the same result:

Make the reliability of the voter rolls the subject of doubt.

That is important.

The message does not necessarily need to convince people of one specific false claim.

Sometimes it only needs to make the underlying institution seem questionable.

That is why narrative families matter.

The exact wording can change.

The underlying objective can remain remarkably stable.

The important signal may be what does not change

If automated systems can vary wording, timing and even personality, then those things become weaker evidence on their own.

The better question becomes:

What remains consistent underneath the variation?

That could include:

• The same narrative family
• The same originating sources
• The same account clusters
• The same domains or links
• The same order of amplification
• The same accounts appearing early
• The same accounts repeatedly appearing together
• The same narratives appearing around particular moments in the election calendar

That is where the structure begins to emerge.

Recurring accounts are only the beginning

Today’s Lookout tells us that 18 accounts appeared in more than one “Doubt the Rolls” wave.

That is useful.

But the next question is more important:

Are the same accounts appearing together?

Imagine four waves:

Wave 1 includes Accounts A, B, C, D and E.

Wave 2 includes A, B, C, F and G.

Wave 3 includes A, B, C, D and H.

Wave 4 includes A, B, C, E and I.

The important finding is no longer simply that several accounts appeared more than once.

It is that A, B and C repeatedly appear together.

That begins to reveal a network.

Then we can ask:

Do they react within similar time windows?

Do they amplify the same source?

Do they repeatedly arrive early in the spread of the story?

Do they behave this way across different rumor families?

That is much stronger evidence than post volume alone.

Humans can coordinate, too

This is where we have to be careful.

Coordination is not the same as automation.

Forty real people can agree in a private channel to post the same message at 8 a.m.

That activity may produce:

• Similar timing
• Shared links
• Common language
• Recurring account clusters

Every account may still belong to a real person.

So the system should not collapse everything into a single question:

Is this a bot?

There are at least two separate questions:

Is this activity coordinated?

and

Does the behavior show signs of automation?

Those findings should be reported separately.

That gives us a much stronger standard.

Instead of saying:

“We found a botnet.”

we might say:

“This wave shows strong indicators of coordinated amplification.”

And, where the evidence supports it:

“The network also displays behavioral patterns consistent with automation.”

That is more precise.

Machines can hide inside ordinary-looking accounts

There is another reason account-level bot detection is becoming less useful.

An account can look completely ordinary most of the time.

It posts about football.

Weather.

Music.

Local news.

Politics.

Food.

Then something changes when a particular election narrative appears.

Every time a “Doubt the Rolls” wave begins, the account:

• Appears quickly
• Amplifies the same small group of accounts
• Uses the same few source domains
• Joins the same recurring cluster

Its ordinary activity looks noisy.

Its mission activity looks regular.

That difference may be one of the strongest signals available to us.

The question is no longer simply:

What does this account normally look like?

It becomes:

How does this account behave when a specific narrative appears?

That is behavioral discontinuity.

The first hour may matter more than the biggest hour

There is another complication.

Automation does not have to carry a rumor forever.

It may only need to get the story moving.

A wave could develop like this:

Automated seed → automated amplification → real influencer notices → real people engage → organic spread

By the time the story becomes large, most of the people sharing it may be completely real.

They may have no connection to whoever or whatever first pushed the claim.

They may simply believe it.

That means looking only at the largest stage of a rumor can obscure how it started.

The first hour may be more revealing than the biggest hour.

Who posted first?

Who reacted first?

Which accounts appeared immediately?

What sources were shared?

How quickly did the claim move from one part of the network to another?

Those questions help distinguish origins from later popularity.

That changes what we should measure

The Lookout already gives us important information:

• Narrative family
• Number of waves
• Time between waves
• Recurring accounts
• Election-calendar context

The next layer should examine how the network behaves.

Time synchronization

How quickly do accounts join a wave?

Twenty accounts posting over two days is very different from 20 accounts appearing within six minutes.

Content similarity

Are the posts exact copies?

Near copies?

Paraphrases?

Or completely different statements advancing the same underlying narrative?

Source concentration

Are dozens of accounts using dozens of independent sources?

Or do most of them eventually trace back to the same obscure source?

Recurring account clusters

Which accounts repeatedly appear together?

This is more useful than simply knowing how many accounts appeared more than once.

Reaction latency

How quickly does each account respond after an apparent source appears?

The same accounts repeatedly reacting unusually quickly across multiple waves deserves attention.

Behavioral discontinuity

Does an account behave differently when a specific election narrative appears?

This may help separate everyday activity from mission-oriented activity.

The election calendar gives us another layer

Timing matters because different narratives can become more useful at different stages of an election.

Michigan begins early voting Oct. 24.

Early voting ends Nov. 1.

Election Day is Nov. 3.

County canvass deadlines arrive Nov. 17.

The state canvass deadline follows Nov. 23.

Before and during voting, rumors may focus on:

• Voter eligibility
• Polling places
• Intimidation
• Whether voting is safe

Around Election Day, attention can shift toward:

• Voting machines
• Tabulation
• Ballot counting

After Election Day, narratives may focus more heavily on:

• Election officials
• Canvassing
• Certification
• Whether the result should be trusted

That does not mean those narratives will appear.

It gives us a way to evaluate timing when they do.

Not every rumor is disinformation

This distinction is essential.

Someone can repeat something false because they believe it.

Someone can misunderstand an election rule.

Someone can share a claim without knowing where it originated.

A model can classify something incorrectly.

And a large group of people can independently arrive at the same concern.

That is why this system should not treat suspicious-looking behavior as proof of intent.

The Lookout identifies what is happening.

The Outlook asks what patterns might explain it.

Human review remains necessary between those two.

The deeper problem is no longer human versus machine

The old question was:

Is this a real person or a bot?

That may be becoming the wrong question.

A better set of questions is:

What story keeps returning?

Which accounts repeatedly carry it?

Which accounts repeatedly appear together?

How quickly do they react?

Where does the material originate?

What happens during the first hour?

What changes when a specific narrative appears?

And, most importantly:

What remains regular underneath the noise?

The surface can be randomized.

The language can be rewritten.

The accounts can look different.

The timing can be varied.

But an operation still has to accomplish something.

It still has to move a story.

It still has to reach people.

It still has to reinforce a belief.

That operational objective may be much harder to disguise than the posts themselves.

Today’s Lookout shows us the waves.

The Outlook should now begin looking beneath them.